[zeromq-dev] Clever uses of XREP/XREQ

[Matt Weinstein]

True.

I think the important point is that _as a platform_ ØMQ makes it much  
easier to experiment with new topologies like this.

If these patterns become important, they will be logical to integrate  
(or layer).

On Jul 24, 2010, at 3:19 AM, Martin Sustrik wrote:

> Brian,
>
>> We are using XREQ/XREP a lot these days and one of our developers  
>> came
>> across some interesting uses of them.  I wanted to see if these uses
>> were official supported, see if others were doing these types of
>> things and point out these interesting cases to others.
>
> ...
>
>
>> We think these are pretty clever uses of these socket types.  Some  
>> questions:
>>
>> * Is this supported behavior (please say yes) or might this go away  
>> in
>> the future?
>> * Are there any issues to be aware of that we might run into in this
>> configuration in a high load context?
>> * Anyone else using these or other sockets in creative ways?
>
> To get a clear idea of what's going on you have to take a bit  
> different
> point of view: Think of XREQ/XREP as IP and REQ/REP as TCP. Basically,
> two different layers of the stack.
>
> Now image a person used to TCP. At some point he discovers IP and
> becomes enthusiastic about it: Wow! You can send packets to any  
> address,
> not just to the peer! Wow! You can get a packet from any peer without
> need to listen/accept new connections! etc.
>
> The above is true but there's a couple of caveats:
>
> 1. If you need any of TCP features (realiability, congestion control)
> you have to implement it yourself.
> 2. If you want to interoperate with an application using TCP, you  
> won't
> be able to (unless you reimplement TCP in your app).
>
> Same applies to XREQ/XREP. It's a low-level protocol. You've built
> couple of protocols on top of it (one-request-many-replies,
> identify-on-connect). In theory these can be made to alternative
> upper-layer protocols (messaging patterns, socket types) at the same
> level as REQ/REP.
>
> If you want to go that way, the questions to answer are: Are the
> semantics of the pattern coherent? Does it work as expected in all
> contexts (different transports, topologies etc.) Does it scale? Etc.
>
> Martin
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev